I am too busy for preparing the on-going interview and have no time to finish the capstone. So now I do a quick review and then drop out of the nanodegree. Dropout, as a powerful technique in deep neural network, is also useful for me to make real-life decision. If you are learning too hard, you may be overfit. It is time to step back and reflect.
As my previous post said, the deep learning section in this nano degree is poorly designed. I got stuck here in the last 2 months and made very slow progress. Deep learning really takes time and effort.
Step 4 of digit_recognition is a real challenge. It makes me realize the real-world scenario is much more complicated. I couldn’t figure out how to do a localizer that can deal with different size of inputs.
And it is my first time to use the 1:1 appointment via Zoom. However, the tutor seems not well prepared. I spend 20 mins getting him understand the difficulty of the project. At last, we look to the Forum for solution: https://discussions.udacity.com/t/tips-for-svhn-project-with-bounding-boxes/219969. But after that, I became super busy with a data scientist opportunity and have no time to go back.
The capstone project is kind of DIY thing. You learn by exploring yourself.
Below are a few suggested problem areas you could explore if you are unsure what your passion is:
Review: is this machine learning ND worth it?
The short answer is: Yes!
Of course, Udacity can do a better job. The point is l learned a lot at my own pace and get prompt feedback. I am pretty confident to talk about machine learning with other colleagues. And I know how to improve myself: keep practicing on the real-world dataset and keep sharing what I learn.
By the way, I didn’t realize that nanodegree is a trademark of Udacity, who applied it in 2004